Thymic tissue from 53 immunologically healthy children aged between 1 day and 11 years was obtained from children undergoing corrective cardiac surgery at the University Children’s Hospital Zurich. In addition, 2 thymi from patients with immunological phenotypes were obtained: 1 patient with myasthenia gravis (15 years old, female, high-dose steroid treatment) and 1 patient with 22q11.2 microdeletion syndrome (14 days old, female, no medication with known influence on thymic cell populations). All tissue was obtained in accordance with the declaration of Helsinki and in adherence to guidelines from the local ethical commission (No. 2017-00312). In total, 55 thymi were used in this study.

Thymic tissue was stored in phosphate-buffered saline (PBS) and processed within four hours. For thymocyte preparations, a small fragment of cleaned thymic tissue was placed between two pieces of 100 μm nylon mesh, and carefully rubbed with the bottom of a syringe. To obtain stromal cells, fragments were digested in PBS with Liberase TM 0.2 mg/ml (Roche) and DNAse I 0.02 mg/ml (Roche) at 37°C under repeated pipetting for 35minutes and filtered through a 100 µm nylon mesh.

Antigen presenting cells (APCs) were enriched byPercoll density gradient (25) centrifugation. In brief, a Percoll (GE Healthcare) solution (final density 1.07 g/ml with physiological osmolarity) was distributed to 2 centrifuge tubes, mixed with 25x10⁸ cells, overlayed with cold buffer (AutoMACS Running buffer, Miltenyi), and centrifuged at 3500xg for 20 minutes at 4°C with decreased acceleration and deceleration.

Samples were stained with fixable viability dye eFluor506 (BioLegend) The following fluorochrome-conjugated antibodies were used: BioLegend: CD11b (M1/70), CD155 (SKIL4), CD165 (SN2(N6-D11), CD1a (HI149), CD200 (OX-104), CD205 (HD-30), CD27 (O323), CD3 (HIT3a), CD31 (WM59), CD4 (OKT4), CD45 (2D1), CD49e (MKI-SAM-1), CD49f (GoH3), E-Cadherin (67A4), EpCAM (9C4), HLA-DR (L243), podoplanin (NC-08); Miltenyi: CD11c (REA618), CD8 (REA734). UEA-1 (Sigma-Aldrich L9006) was used at 100 μg/ml. Samples were fixed with Fixation Buffer (BioLegend 420801).

The Foxp3/transcription buffer kit (eBioscience 00-5523-00) was used after surface marker staining. Antibodies: cytokeratin 8 (abcam, EP1628Y, 1:1000), cytokeratin 13 (abcam, EPR3671, 1:2000), cytokeratin 14 (ThermoFisherScientific, LL002, 1:100), cytokeratin 15 AlexaFluor555 (abcam, EPR1614Y, 1:500).

For the Legend Screen (BioLegend 700001), CD45⁺ cells were partially depleted from digested tissue (15x10⁸ cells) anti-human CD45 beads (Miltenyi 130-045-801) and mixed with 7x10⁷ undepleted cells. Cells were pre-mixed with antibodies against CD45 (BioLegend, 2D1) and EpCAM (BioLegend, 9C4).

All flow cytometry data were acquired on a BD LSR Fortessa (BD Biosciences). Data were analysed using FlowJo (version 10).

For fluorescence-activated cell sorting (FACS), thymic stromal cells were enriched by Percoll-density gradient centrifugation (25) followed by depletion of CD45⁺ cells from the APC-enriched cell suspension (5x10⁸ cells) with anti-human CD45 beads (Miltenyi 130-045-801).subsequently stained with fluorochrome-conjugated antibodies. This stepwise enrichment lead to an approx. 200-fold proportional enrichment of TECs, fibroblasts and endothelial cells ( Figures S2K, L ). Propidium iodide was added to assess cell viability. 1-2x10⁵ live cells were sorted on a FACSAria fusion (BD Biosciences) and resuspended in RLT lysis buffer (Qiagen).

Ribonucleic acid (RNA) was extracted with the Qiagen RNeasy micro kit (Qiagen, 74004). Reverse transcription was performed with theQiagen QuantiNova Reverse Transcription kit (Qiagen, 205413). qPCR was performed with the Qiagen QuantiNova qPCR master mix (208154) (on a 900HT Fast Real-Time PCR System (Applied Biosystems). Primers were used at a concentration of 10 µM: AIRE forward 5’-CAA GGA TGA CAC TGC CAG TC-3’, reverse 5’-TGC TCT GGA TGG CCC ACT G-3’, reference genes: GAPDH forward 5’-CTT CAA CAG CGA CAC CCA CT-3’, reverse 5’-TGCTGTAGCCAAATTCGTTGTC-3’, HPRT-1 forward 5’-CCT GGC GTC GTG ATT AGT G-3’, reverse 5’-TCG AGC AAG ACG TTC AGT CC-3’, RPLP13 forward 5’-CGG ACC GTG CGA GGT ATG CT-3’, reverse 5’-AGC AGG AAC CAC CAT CCG CT-3’. Results were analysed using the 2^-ddCt method (26).

The SMARTer^® Stranded Total RNA-Seq Kit v2 -Pico Input Mammalian (A Takara Bio Company, California, USA) was used for total RNA samples (0.25–10 ngThe quality and quantity of the enriched libraries were validated using a Qubit^® (1.0) Fluorometer and the Tapestation (Agilent, Waldbronn, Germany). The libraries were normalized to 10nM in Tris-Cl 10 mM, pH8.5 with 0.1% Tween 20.

After library quantification, libraries were prepared for loading accordingly to the NovaSeq workflow with the NovaSeq 6000 Reagent Kit (Illumina, Catalog No. 20012865).

Cluster generation and sequencing were performed on a NovaSeq 6000 System with a run configuration of single end 100bp.

The raw reads were first cleaned by removing adapter sequences, trimming low quality ends, and filtering reads with low quality (phred quality <20) using Trimmomatic (Version 0.36) (27). Sequence pseudo alignment of the resulting high-quality reads to the Human reference genome (build GRCh38.p10, Ensembl release 91) and quantification of gene level expression was carried out using Kallisto (Version 0.44) (28).

The normalized signal was internally calculated by EdgeR using the TMM method (29).

Raw data is publicly accessible at the European Nucleotide Archive database under accession number PRJEB39649.

For statistical comparison a QL (Quasi-likelihood) F-Test was used to calculate p values. The pValue adjustment was performed with the Benjamini-Hochberg method (30). The log2 Ratio was calculated based on TMM normalised counts using the sample group mean values.

Thymic tissue (fresh or fixed with 4% paraformaldehyde) was embedded in OCT (Biosystems Switzerland, 81-0771-00) and snap-frozen in methanol/dry ice Tissue blocks were cut into 8 μm sections onto Superfrost Plus Slides (Thermo Scientific).

Samples were fixed with ice-cold methanol for 4 minutes (AIRE, cytokeratins) or aceton (collagen IV α5, α6) or 4% paraformaldehyde followed by permeabilization with 0.25% Triton-X100 (CD200, collagen IV α3, FN1 and CD49f). For pdpn, heat-mediated antigen retrieval was performed at pH 6 in 20 mM Citrate buffer for 5 minutes.

All samples were blocked prior to the staining procedure with PBST + 5% fetal calf serum Fluorochrome conjugated antibodies were: cytokeratin 8 AlexaFluor488 (abcam, EP1628Y, 1:100), cytokeratin 5 AlexaFluor647 (abcam, EP1601Y, 1:100), cytokeratin 13 AlexaFluor647 (abcam, EPR3671, 1:100), cytokeratin 14 FITC (invitrogen, LL002, 1:50), cytokeratin 15 AlexaFluor555 (abcam, 1:100). Unconjugated antibodies were: AIRE (abcam, ab78965 goat polyclonal, 1:100), pdpn (abcam, 18H5 mouse monoclonal, 1:100), CD200 (Invitrogen, PA5-47375 goat polyclonal, 1:50), COL4A3 (abcam, ab223227, goat polyclonal, 1:25), CD49f (Biolegend, GoH3, rat monoclonal, 1:25), FN1 (abcam, IST-9, mouse monoclonal, ab6328). UEA-1 (Sigma-Aldrich L9006) was used at 50 μg/ml.

Secondary antibodies: (donkey anti-goat IgG H&L Alexa Fluor555, abcam, ab150134; goat anti-rabbit IgG H&L, Alexa Fluor647, Invitrogen, A21244; goat anti-rat IgG H+L, Alexa Fluor488, Invitrogen, A11006; goat anti-mouse IgG1, AlexaFluor647, Invitrogen, A21240) diluted 1:1000. 4′,6-diamidino-2-phenylindole (DAPI) at 5 μg/ml was added for nuclei and slides were mounted with Prolong Diamond Antifade Mountant (Molecular Probes P36961).

Images were acquired on a Nikon Ti2-E with a Plan Apo λ 20x Ph2 DM (numerical aperture 0.75) and a Plan Fluor 40x Oil DIC H N2 (numerical aperture 1.3) objective with a DS-Q2 camera and the NIS Elements Advanced Research software. Higher resolution images were acquired on a Leica SP8 inverse confocal laser scanning microscope with a 63x glycerin immersion objective [HCX PL APO 37°C CS2 (numerical aperture 1.3)] at room temperature with the Leica LAS X software.

Images were processed in ImageJ2 (31).

Data plots were generated using the ggplot2 package (32) in R (33) in combination with the tidyr package (34). The Venn Diagram was created using the package “VennDiagram” (35), the heatmap was created using the package “heatmap3” (36).

EpCAM, CD249 and UEA-1 clearly identify murine TECs and their subsets within the CD45^- stromal cells (2). In humans, EpCAM expression is variable with cTECs generally expressing lower levels than mTECs (18, 37). Using cytokeratin (ck)8 or ck5 as specific markers of all human TECs ( S1A and S1B Figure ) (20, 38), we found that a subset of ck+ cells exhibited very low EpCAM surface levels, especially in thymic cells from some donors ( Figure 1A ), indicating that EpCAM expression alone is insufficient to clearly distinguish human TECs with low-level EpCAM expression from non-TEC stromal cells ( S1C Figure ). Thus, we subjected human thymic stromal cell preparations to an FC surface marker screening (332 markers) aiming at identifying more suitable surface markers for the comprehensive identification of human TEC. Within the CD45^- population we assessed the MFI of the interrogated markers, ranked as ratio between the EpCAM^high/int vs. EpCAM^- population ( Figures 1B, C ). NC-08, an antibody recognizing pdpn, emerged as candidate discriminating TECs from the other thymic cell types in conjunction with EpCAM. The comprehensive coverage of all human TECs using EpCAM and pdpn was confirmed by FC using intracellular staining for ck8 ( S1D Figure , S1E Figure ).

Human cTECs are commonly identified using antibodies against CD205 (2) or the CDR2-antibody, recognizing a yet unknown epitope (39). UEA-1 was reported to label human mTEC (20). We found that in multiple human thymus samples, the combination of UEA-1 and CD205 left a substantial fraction of TECs unstained ( S1G Figure ), and while UEA-1 staining was exclusively found in the medulla, it failed to label all mTECs ( S1F Figure , S1G Figure ). EpCAM expression is reportedly higher on human mTECs (EpCAM^high) than cTECs (EpCAM^low) (18, 37). Aware of the caveat that we may omit some EpCAM^low cTECs in our analysis, we used the data from the above FC screening assay and interrogated EpCAM^high vs EpCAM^low TECs ( S2A Figure ) for more distinct yet comprehensive markers. Ranking the ratio of the MFI assessed in EpCAM^high vs EpCAM^low cells identified markers with high discriminatory capacity between the two populations ( Figure 2A ). The top 30 markers from each end of this spectrum were further evaluated for their absolute MFI, as high expression levels are potentially more suitable for FC ( Figures 2B, C ). Potential markers were confirmed by visual inspection of the FC data based on uniformity of the subpopulations and clear-cut discrimination between them, which led us to exclude CD54 and CD82 ( S2B Figure ). SSEA-4 and SSEA-5 were dismissed as these are glycolipids rather than surface proteins and thus not easily correlated with the expression of particular genes. An additional set of 4 thymi was used to evaluate the best candidates (CD200, CD342, CD155 for mTECs, CD49f, CD166, CD49e, CD51, CD165 for cTECs) for inter-individual reproducibility ( S2C Figure ). Based on these criteria, antibodies reactive to CD200 (on mTEC/EpCAM^high) and CD49f (on cTEC/EpCAM^low) respectively proved to be most effective due to high expression levels and low inter-individual variability ( Figure 2D and S2C Figure ).

Immunofluorescence analysis of thymus cryosections revealed exclusive medullary staining for CD200 ( Figure 2E ), but not restricted to mTECs. In fact, FB, EC and a small fraction of B cells express CD200 as assessed by FC ( S2E Figure ). The corresponding receptor, CD200R, was mainly found on B cells, macrophages and dendritic cells ( S2F Figure ). Prominent reactivity to CD49f was detected on intrathymic vascular structures by immunofluorescence analysis ( S2D Figure ), rendering it impractical as a discriminatory marker for cTECs in IF analysis. This is in accordance with FC data ( S2J Figure ). However, vascular endothelial cells are easily discriminated from pdpn⁺ cTECs based on pdpn and EpCAM expression ( S2K Figure ). Lymphatic endothelial cells co-express high levels of CD49f and CD200 as well as CD31 and pdpn, allowing their clear distinction from TECs ( S2I, J Figure ).

We thus propose the following strategy for the identification of TEC: pdpn^high/intEpCAM^high/int/low for TECs, with further subdivision of CD49f⁺CD200^- for cTECs and CD49f^int/lowCD200⁺ for mTECs. Interestingly, CD45 proved to be dispensable for our gating strategy ( S1D Figure and data not shown).

Gene expression analysis of AIRE and FOXN1 within accordingly FC-sorted TEC subpopulations and a CD45^-EpCAM^- control cell population demonstrated exclusive expression of AIRE in the mTEC subset, whereas FOXN1 expression was found in both subsets but not in the control cells ( S2G Figure , S2H Figure ).

In view of a more profound validation of the proposed gating strategy we sought to gain a detailed understanding of the transcriptional activity of the individual stromal cell subpopulations using RNA-sequencing of bulk sorted subpopulations. To this end we sorted cTECs and mTECs according to our proposed gating strategy (6 individuals) as well as CD45^-CD31⁺pdpn^- vascular endothelial cells (EC) and CD45^-HLA-DR^-pdpn^int fibroblasts (FB) (4 individuals). Identically sorted cell populations grouped together when unsupervised clustering of normalized gene expression data was applied. Additionally, TEC subpopulations distinctly separated from non-epithelial cell subsets ( Figure 3A ).

The identity of each subset was assessed interrogating characteristic gene expression profiles. In accordance with published data FOXN1 (cTEC > mTEC) and KRT8 (ck8) expression was increased in TECs, whereas CDH5 (VE-Cadherin) and COL1A2 were found exclusively in ECs and FBs respectively, thus confirming the identity of the subpopulations sorted according to the novel gating strategy ( Figure 3B ) (16, 20). Gene expression levels of the markers used for FC identification and sorting, EPCAM (EpCAM), PDPN (pdpn), ITGA6 (CD49f) and CD200 (CD200) were distributed according to the expected pattern ( S3A Figure ).

Furthermore, markers associated with thymocytes, CD7 (CD7) and PTPRC (CD45) (16), were only detectable at very low levels in all four cell populations ( S3B Figure ), suggesting minimal contamination thereof. According to recent data from scRNA-sequencing there is substantial heterogeneity with both cortical and medullary TECs (16, 18, 40). We therefore assessed our dataset from bulk-sorted cTEC- and mTEC subpopulations for signatures associated with proposed subsets. The distribution of genes associated with tuft-like cells (POU2F3, IL25), neuroendocrine cells (NEUROD1, CHGA), myoid cells (MYOG, MYOD1, BEX1), corneocyte-like cells (FXYD3, LYPD2, IL1RN), ciliated cells (SOX2, FOXJ1), immature TECs/cTECs (CTGF, ZBED2) or mTECs (ASCL1, INSM1) were found to be upregulated in the respective TEC population ( S3C Figure ).

Medullary TECs contribute to self-tolerance through the expression of TRAs at very low levels (2, 37). Furthermore, each TRA is only expressed in about 1-3% of all mTECs (37). We therefore looked for low-level unique gene expression across all samples. All genes expressed at >0.5 fragments per kilobase of transcript per million mapped reads (FPKM) in at least 3 out of 6 (cTEC and mTEC) or 2 out of 4 (FB and EC) samples were checked for their expression across all four cell populations. Housekeeping genes -as defined by the human protein atlas- were excluded ( S3D Figure ). The mTEC subset demonstrated by far the highest number of exclusive transcripts (841), compared to 128 in cTECs, 227 in FB and 104 in EC. Comparing these 841 genes to the human protein atlas tissue specific gene database most of them (796) could be assigned to one of the tissue-specificity categories ( S3E Figure ). The 263 genes that fell into the “tissue enriched” category contained well-known TRAs such as Insulin (not shown). The highest number of genes could be assigned to brain (41), followed by testis (42). Other organs with a high number of TRAs (liver, kidney, pancreas) were also represented ( S3F Figure ), confirming a hallmark feature of this population sorted according to the proposed strategy.

Thymic stromal cell development and differentiation is guided by a number of transcriptional regulators, which control the expression of factors involved in cell-type specific growth and function ( S1 Table ). We therefore analyzed the expression levels of genes with known and suspected involvement in transcriptional regulation of thymus development, mTEC maturation/function, antigen presentation, cTEC function, cell-cell signaling from mouse or human studies ( Figure 3C and S1 Table ).

In line with previous findings the expression of transcription factors and regulators associated with thymic development were up-regulated in TEC subpopulations with the exception of TBX1 and PAX3, which are involved in patterning of the pharyngeal pouches. Whereas AIRE, ASCL1, EHF, POU2F3 and SPIB expression were restricted to mTEC, neither the transcription factor FEZF2 nor the transcriptional regulators SIRT1 or PRDM1 demonstrated specific up-regulation in this bulk-sorted population. In accordance with published observations (20, 43), expression of the mTEC-associated tight junction proteins CLDN3 and CLDN4 was restricted to the mTEC population.

The analysis related to phenotypic and functional characteristics of fibroblast and endothelial cells demonstrated high levels of PECAM1 (CD31) and CDH5 (VE-cadherin) expression in endothelial cells as expected (16). The CD45^-EpCAM^-HLA-DR^-pdpn^int population uniquely expressed fibroblast-associated markers (including FAP, S100A4, PDGFRA) and contains the markers described for the populations Fb1 (intrathymic FB) (COLEC11, GDF10, ALDH1A2) and Fb2 (capsular FB) (FBN1, PI16, SEMA3D) (18). Notably, the fibroblast population distinctively expressed high levels of FGF7 and FGF10, known for their contribution to TEC proliferation and differentiation (9) ( Figure 3D ), whereas the corresponding receptor (FGFR2) is highly upregulated in both cTECs and mTECs ( Figure 3C ). Interestingly, RNA levels of molecules involved in immunomodulatory function of fibroblastic reticular cells (FRCs) in the periphery (such as PTGS2, IDO1 and IDO2 and NOS2 (44) are more pronounced in TECs.

In murine studies, co-stimulatory molecules (such as CD80) are commonly used in conjunction with MHCII expression to identify mTEC maturation stages (42). However, on human TECs we were unable to detect protein expression of CD80, CD86 or OX40L in our FC surface marker screen ( S3G Figure ). Another costimulatory molecule, CD40, was detected at high levels on both cTECs and mTECs ( S3G Figure ), although literature suggests an exclusive expression on human mTECs (45). Tissue preparation may affect epitope availability, although expression of each of these molecules was detectable on CD45⁺ cells (data not shown). In accordance with our surface marker screen, transcript levels of CD80, CD86, and OX40L were very low. Yet, mRNA and protein levels for CD40 were found to a similar extent on both TEC subsets ( S3G Figure ).

Cytokeratins (cks) are critical for epithelial cell structure and function. Our dataset showed robust differential expression for ck13 ( S4A Figure ), ck14 ( S4B Figure ) and ck15 ( S4C Figure ). The restriction of ck13 expression to mTECs was confirmed by FC and IF, yet found to be independent of AIRE expression. On the other hand, ck14 and ck15 were found in almost all TECs by FC and IF ( S4 Figure and data not shown).

For their growth and migration developing thymocytes rely on specialized and highly organized ECM produced by stromal cells (46). Even though the three-dimensional network structure of TECs is devoid of a classical basement membrane, collagen IV, fibronectin and laminin, which are the principal components of basement membranes are reportedly part of the thymic ECM (47, 48). The most significant contribution to gene transcripts associated with the GO term “extracellular matrix” came from FB and EC, and included collagen IV subtypes COL4A1 and A2 ( S5A Figure , S5B Figure ). Interestingly, the α3 to α6 chains of collagen IV ranked among the most differentially regulated genes in cTECs versus mTECs ( Figure 4A ). Thus, while the contribution of TECs to this gene set seems to be limited, it is highly subset-specific, with COL4A3 and COL4A4 expression being significantly higher in mTECs, whereas COL4A5 and COL4A6 are expressed predominantly in cTECs ( Figure 4A ). Corresponding protein expression was confirmed by IF ( Figure 4B and S5C Figure , S5D Figure ). Gene expression of FN1 (fibronectin) was most prominently expressed in TECs in comparison to other stromal cells ( S5A Figure , S5B Figure ), yet detection of the protein was found mainly in the medulla ( Figure 4C ) in close proximity to mTECs ( S5E Figure ). Collagen IV fibres of the basal membrane serve as structural framework for epithelial cells and other cell types with distinct integrins linking the cytoskeleton to specific collagen heterotrimers (49). In particular, α_vβ₃ and α_vβ₅ integrins bind to the α3α4α5 or the α5α5α6 heterotrimer of collagen IV (49). As the prevalence of different collagen IV heterotrimers seemed to be specific for either thymic cortex or medulla, we investigated the presence of both integrin dimers on several thymic cell types. Whereas thymocytes were devoid of either integrin expression, a large fraction of stromal cells including dendritic cells (DCs), macrophages (MPs) and TEC subsets expressed one or both integrins ( Figures 4D, E ).

Thymic pathology can be observed in a number of inborn or acquired conditions such as microdeletion 22q11.2 syndrome, myasthenia gravis (MG), or high-dose steroid treatment (50–52). We therefore assessed the applicability of our TEC identification strategy on thymic biopsies from patients with these entities. The established gating strategy reliably identified TECs and their subpopulations in these patients. It also revealed significant alterations of MHC class II expression in both conditions compared to an immunologically healthy control, with markedly lower (MG patient) or very low (22q11.2 microdeletion patient) HLA-DR expression on both cTECs and mTECs ( S6A Figure , S6C Figure , S6E Figure ). The cytokeratin distribution in the thymus biopsy from the patient with MG was severely altered, with disruption of the epithelial network and aberrant cortico-medullary demarcation compared to a healthy control ( S6B Figure , S6D Figure ). The thymus from the patient with microdeletion 22q11.2 on the other hand demonstrated normal thymic architecture and inconspicuous distribution of cytokeratins ( S6F Figure ).

Having established a reliable and comprehensive identification strategy for human TECs enabled us to quantify TECs by FC in a total of 31 thymi from donors age 1 day to 12 months. The median fraction of TECs in human thymus decreased – though not reaching statistical significance - from 0.067% in the 0-3 months old infants, to 0.033% in patients aged 3-6 months, to 0.016% in the 6-12 months age group ( Figure 5A ). Dendritic cell frequency (CD45⁺HLA-DR⁺CD11c⁺CD11b⁺) and LEC (CD45-CD31+pdpn+) followed the same trend, whereas the proportion of macrophages (CD45⁺HLA-DR⁺CD11c^-CD11b⁺), EC (CD45-CD31+pdpn-) remained constant in these age groups (S7A-E). Averaged across the age groups, we found median frequencies of 0.044% (MP), 0.057% (DC), 0.06% (FB), 0.061% (EC) and 0.0013% (LEC) in native thymi. Interestingly, male infants (age 0-3 months) demonstrated a significantly decreased cTEC/mTEC ratio due to a disproportionate decrease of the cTECs in comparison to mTEC frequency in this group of patients ( Figures 5B–E ). Thymocyte development and TECs are connected in a mutual regulatory network, a.k.a. thymic crosstalk (11). Hence, we sought to identify alterations in the thymocyte compartment associated with the observed changes in the TEC compartment. We found a significant inverse correlation when comparing the ratio of mature (CD3^highCD1a^low) CD4- to CD8-single-positive thymocytes with the cTEC/mTEC ratio ( Figure 5C ). The frequency of mature (CD3^highCD1a^low) CD8-single-positive thymocytes showed a trend for increase with the surge of the cTEC/mTEC ratio, whereas the opposite was seen for the mature CD4-single-positive counterpart ( Figures 5F, G ).

In the present study we identified surface markers useful in the comprehensive flow-cytometric identification of human TECs (pdpn in conjunction with EpCAM) as well as their reliable partition into cortical and medullary subpopulations (CD49f and CD200). The combination of these markers provided unambiguous positive identification of all TECs, even those with low EpCAM expression, and demonstrated high inter-individual reproducibility.

Previous strategies to analyse human thymic tissue were mostly based on histological analysis (20, 23), or on thymocyte and peripheral T cell populations subpopulations as indirect markers (53). By contrast, FC provides a simple and relatively fast method for tissue analysis. We thus took advantage of our approach to analyse 31 thymi from children in their first year of life. We found a pronounced reduction in cTEC/mTEC ratio in males younger than 3 months of age compared to age-matched females, which resulted from a relative lack of cTECs rather than an increased proportion of mTECs. Androgens are known to negatively impact on TEC proliferation and regeneration (52, 54), and especially impair cTEC regeneration (54, 55), whereas estrogens have a negative effect on mTECs (56).

In humans, minipuberty, a physiological post-natal surge in androgens in male infants younger than 6 months (57), has been correlated with sex differences in thymus function in human infants (58, 59). Thus, our findings corroborate these reports on the effects of gonadal steroids on TEC composition, suggesting a negative impact of androgens on the frequency of cTECs, which might explain previously observed relative increases in mTECs and AIRE expression in human male infants.

Peripheral T cell function is crucial for efficient immune response and prevention of autoimmunity. As these might be directly related to thymic output and thymocyte development (60), we were interested whether the androgen-related change in thymic stroma composition could have an influence on thymocyte populations. We found that the cTEC/mTEC ratio was negatively correlated to the ratio between mature CD4 and CD8 SP thymocytes. As CD8+ thymocytes require longer cortical dwell time, these cells might benefit from a larger thymic cortex (61). Studies in mice demonstrated that different proteolytic enzymes are responsible for the generation of either CD4 and CD8 thymocytes, highlighting the importance of cTECs for thymocyte development (62, 63). Along these lines, patients with thymoma of preferential cortical phenotype demonstrate an increased proportion of CD8+ T cells in the peripheral blood (64). On the other hand, androgens may have a direct effect on thymocyte development (65).

Our proposed approach to comprehensively quantify human TEC subpopulations by FC may contribute to a better understanding of stromal changes in diseases associated with thymic defects, as demonstrated in samples derived from patients with microdeletion 22q11.2 and myasthenia gravis (50, 51). Also, it may easily be applied to the evaluation of thymic grafts pre- and post-transplantation and thus complement currently used histological methods (24).

The thymic expression pattern of CD200 with its restriction to the medulla and its high expression on all mTECs raises several questions as to its potential contribution to human T-cell development. In mice, this differential expression pattern was not observed (data not shown). CD200 was initially discovered as a T-cell co-stimulatory molecule, but was later shown to induce an immunomodulatory program in antigen-presenting cells and is thus considered an immune checkpoint molecule (41). CD200 expression on various hematological and solid tumors may contribute to immune evasion partly due to induction of regulatory T cells (66–68). Although a direct effect of CD200 expressed by mTECs on thymocytes is unlikely – the corresponding receptor is lacking on thymocytes (data not shown) – it may induce a modulatory effect on thymic medullary APCs, some of which express the corresponding receptor (data not shown).

The integrin CD49f has previously been described as a marker for human TECs (69, 70). We found that addition of CD49f to our flow cytometry approach proved valuable for a better discrimination between cTECs and mTECs, as the latter generally show a lower CD49f expression level.

Taking into account the expression of pdpn restricted to an mTEC precursor population at the cortico-medullary junction in mice (71), the generic distribution on human TECs was unanticipated, but suggests significant species-specific differences in its regulation of expression and potential biological function. This evolutionarily conserved glycoprotein is critical for the development of the lymphatic system and lymphoid organs and has been detected in healthy and diseased human thymus previously (72). It contributes to cell adhesion and interacts with soluble and cell-bound proteins (including chemokines) and contributes to the local accumulation and formation of gradients of growth and differentiation factors (73). In murine lymph nodes, pdpn orchestrates the migration of DCs and tissue homeostasis directly (through interaction with CLEC2) and indirectly (through gradient formation) (74). By sequestration of the CCR7 ligand CCL21 on murine thymic fibroblasts, pdpn is involved in the formation of regulatory T cells (75). The broad function of pdpn in lymphatic organs suggests that it also plays a vital role in human TEC biology, either by interaction with growth factors or by distribution and function of DCs.

In accordance with numerous studies in mice and humans, transcripts associated with central tolerance (AIRE and TRAs) were upregulated in mTECs. However, we found significant differences to observations derived from murine studies. Neither FEZF2, SIRT1 and HDAC3 were differentially expressed by specific stromal cell subpopulations (see S1 Table for references). FEZF2 was shown to regulate AIRE-independent clusters of TRAs, and both molecules were important for self-tolerance in mice (76). In comparison to other human thymic APC, expression was reported exclusively in mTECs (76, 77). In our study with a focus on non-HSC-derived stromal cells we find that FEZF2 is similarly expressed in all four stromal cell populations, however this might also be due to our bulk sorting and analysis strategy. Taken together, we have identified several novel molecules on human thymic stromal cells that differ from their expression in the mouse counterparts.

Murine mTEC maturation is in parts defined by the expression of co-stimulatory molecules such as CD80, CD86, CD40 and OX40L, which are critical for thymic crosstalk and induction of thymic tolerance in mice. Especially CD80 is often used in conjunction with MHC II to determine the stage of mTEC maturation, but for human thymus, this has not been described so far (11, 42). Neither gene expression nor surface protein expression analysis revealed substantial levels of these molecules on mTECs in our dataset, with the notable exception of CD40, which we also found on FB and EC, although initial evidence from human thymus suggested CD40 expression restricted to mTECs (45). Previous reports of costimulatory molecules in the human thymus have highlighted their expression by non-epithelial APCs and their importance for Treg generation (78–80), indicating that their expression on mTECs might be dispensable and thus has limited suitability to serve as maturation markers in the human organ as opposed to mouse thymus.

Both major TEC subsets exert specialized functions with vastly differing molecular mechanisms. The functional dichotomy in the thymus is underscored by the histological partition into distinct compartments. In polarized epithelia, collagen IV is the hallmark collagen of the basal membrane (49), but the thymus is devoid of a clearly polarized architecture (81). The presence of the collagen IV α3 chain has been described before in human thymus Hassall’s corpuscules, but was considered a TRA (82). Furthermore, collagen IV was detected in the human thymus, but mainly localized to the septum and blood vessels (47). This might be due to strong expression of COL4A1 and COL4A2 by fibroblasts and endothelial cells according to our data. Even though our transcriptomic analysis revealed that in the thymus the bulk of ECM is contributed by fibroblasts and endothelial cells, we found that differing production of collagen IV α-chains and fibronectin by cTECs and mTECs might aid to shape the compartmentalization into thymic microenvironments. Fibronectin expression has been shown in the thymic medulla and as an important factor for thymocyte-TEC adhesion in vitro (47, 83), although the cellular source of fibronectin in the thymus had not been investigated.

Our data suggest that TECs produce a restricted set of ECM components in a highly compartment-specific fashion. Those might serve as adhesion cues for developing thymocytes, which require different adhesion structures during different phases of their development (84). In addition, thymic stromal cells themselves are likely to rely on these structures, as demonstrated here by their expression of integrins specific for collagen IV trimers (49). Whether this interaction provides mere adhesion or is also involved in the commitment and/or maintenance of TEC subsets remains to be clarified. Understanding the role of compartment-specific ECM in thymic function may have important implications in the production of matrices for organoids (46, 85) or in the use of decellularized thymi for the generation of thymic grafts (70).

In lymphatic organs of mice, which are devoid of epithelial cells, FRCs and lymphatic endothelial cells express the highest levels of pdpn (44). Intrathymically, we found the highest levels of pdpn on lymphatic endothelial cells followed by TECs, which prompted us to further investigate overlapping features between fibroblasts and TECs. We found genes characteristic for FRCs (IDO1, IDO2, NOS2) in TECs, and genes characteristic for TEC (IL7, CCL21) in FB. Recent analysis of cultured human TECs also indicated a partly shared gene expression signature between TECs and mesenchymal cells (70). The picture emerging from these studies (reviewed in (6) and our results demonstrate that non-hematopoietic stromal cells (TECs, FB and EC) participate in a complex division of labor with partly overlapping functions similar to what has been described for secondary lymphoid organs (44).

Future studies aimed at investigating redundant and distinctive roles of human thymic stromal cells will aid to define strategies for thymic regeneration and improved graft preparation. Providing detailed phenotypic and transcriptional characterization of the key stromal cell subsets of the human thymus – with a particular focus on TECs - our study significantly contributes to this endeavor.